本研究從國防組織内部控制與内部審核觀點，彙整監察院審計部於90至92會計年 度預算執行期間，有關國防財務(物)審計稽察違失事件，並應用資料探勘技術-Kohonen 類神經網路與C5.0/See5決策樹演算法建立分析模型，找出造成違失事件的原因及資料 各屬性間的關連性，以預防「財務失事」案件發生。 本研究原始資料經正規化後得有效資料共1966筆，利用亂數將資料區分60/40%比 例以進行資料分類模型的訓練與測試，總共萃取出11項具影響財務違失事件的資料屬 性，並針對目標屬性-「退審案件性質」的内涵加以分析，共得到6類27項決策規則。 各項決策規則信心水準，其中達0.7者佔總規則數的70.37%(19/27)，超過0.9者更達 37.03%(10/27);因此，本研究分析模型確實可提供國軍内部審核單位專業人員預測可能 發生「財務違失事件」的高風險單位，進而採取預防措施，增進組織財務效能之有效運作

Basing on an internal control and internal auditing viewpoints for defense organizations, this study collects defense finance auditing violation events from the Ministry of Auditing in the 2001-2003 fiscal year budget execution period. This study applies data mining schemes, Kohonen neuron network and C5.0/See5 decision tree algorithm, to build an analysis model and tries to find the reasons why finance violation events could be caused and the correlations among the attributes of data. The research result could be used to prevent finance violation events. After normalizing the raw data, 1966 records are used to analyze further. All 1966 records are randomly divided into two categories, training and testing, according to a 60/40 ratio. The 11 attributes that would impact finance violation events are extracted. Moreover, analyzing "returned case character" further, 27 rules which can be divided into 6 categories are found. All decision rules have their confidence levels. Among 27 decision rules, there are 19 decision rules (70.37%) have 0.7 confidence level. Moreover, there are 10 decision rules (37.03%) are over 0.9 confidence level. The analysis model proposed in this study would help military internal auditing personnel to predict which organization might cause finance violation events and adopt some prevention measures to improve an organization finance operation performance.